of Engineering. Yours faithfully

We present a brief analysis of robotic programming techniques in the context of service robots, deriving the requirements of such systems to incorporate a high-level planner, a behaviour-based layer responsible for addressing unstructured problems, and machine learning modules for working with specific, complex motions which may involve many degrees of freedom. As the most important requirement for large-scale systems, we see the traceability of observed aspects of behaviour back to the modules causing these aspects. We present an initial prototype developing these ideas in application to the task of navigation in a cluttered, maze-like environment. The prototype uses an expand-able set of event monitors, the states of which regulate which behaviour rules are enabled at any given time. Behaviour rules have both a level and a weight: levels for unconditional precedence over other rules, when such is required, and weights to facilitate an arbitrary selection of behaviours which are at the same level. The sys-tem allows behaviour rules to be added and modified rapidly. Record-and-repla